CN116310352B

CN116310352B - Alzheimer's disease MRI image multi-classification method and device

Info

Publication number: CN116310352B
Application number: CN202310108568.4A
Authority: CN
Inventors: 卢洁; 闫少珍; 於帆
Original assignee: Xuanwu Hospital; Shenzhen Yiwei Medical Technology Co Ltd
Current assignee: Xuanwu Hospital; Shenzhen Yiwei Medical Technology Co Ltd
Priority date: 2023-01-20
Filing date: 2023-01-20
Publication date: 2024-04-12
Anticipated expiration: 2043-01-20
Also published as: CN116310352A

Abstract

The invention relates to the technical field of image classification, and discloses a multi-classification method and device for MRI images of Alzheimer's disease, wherein the multi-classification method comprises the following steps: receiving a brain MRI image of a patient, inputting the brain MRI image into a pre-constructed brain structure identification model to obtain an MRI temporal lobe map, an MRI frontal lobe map, an MRI top lobe map and an MRI occipital lobe map, wherein the brain structure automatic identification model comprises a convolution layer, a feature fusion layer, an attention layer and a YOLO model layer, the area occupation ratio of the MRI temporal lobe map, the MRI frontal lobe map, the MRI top lobe map and the MRI occipital lobe map in the brain MRI image is calculated, and the disease severity level of the Alzheimer disease of the patient is determined according to the area occupation ratio. The invention mainly aims to solve the problem that the prior majority of methods only depend on a single model to identify the brain lobe region of the brain MRI image, so that the error judgment rate of Alzheimer disease is high.

Description

Alzheimer's disease MRI image multi-classification method and device

Technical Field

The invention relates to an MRI image multi-classification method and device for Alzheimer's disease, and belongs to the technical field of image classification.

Background

Alzheimer's Disease (AD) is a degenerative disease of the nervous system that develops with hidden disease progression. Clinically, global dementia is characterized by memory impairment, aphasia, disuse, disrecognition, impairment of visual space skills, executive dysfunction, personality and behavioral changes, and the like. Magnetic resonance imaging (Nuclear Magnetic Resonance Imaging, abbreviated as MRI), also called spin imaging (nuclear magnetic resonance, abbreviated as NMR) is based on the principle of nuclear magnetic resonance, and the position and type of the nuclei constituting the object can be obtained by detecting the electromagnetic waves emitted by the external gradient magnetic field according to the different attenuation of the released energy in different structural environments inside the substance, thereby being able to draw a structural image inside the object.

Clinically, MRI images of Alzheimer's disease show that the MRI images of a patient with mild senile dementia can find that the temporal lobe can be slightly atrophic, and other brain lobe areas can not be obvious; patients with middle-aged senile dementia can see atrophy in the whole brain lobe area, wherein the temporal lobe area is relatively heavy; brain atrophy is very evident in patients with severe senile dementia, with temporal lobe areas and corresponding hippocampal atrophy being more pronounced. Therefore, the brain MRI image is obtained by the MRI technology, and the degree of Alzheimer's disease can be effectively diagnosed.

At present, the diagnosis of the disease course of the Alzheimer disease based on the brain MRI image mainly depends on manual diagnosis or technical diagnosis, wherein the manual diagnosis mainly depends on experienced doctors, and the disease degree of the Alzheimer disease is subjectively judged according to the atrophy degree of each brain leaf in the brain MRI image, and the manual diagnosis has low efficiency but low misdiagnosis probability strictly speaking. The technical diagnosis mainly relies on digital image processing technology or machine learning and deep learning models to perform intelligent identification of each brain lobe area of the brain, and performs area ratio calculation on each identified brain lobe area, so as to estimate whether brain atrophy occurs.

Obviously, the technical diagnosis can obviously improve the diagnosis efficiency of Alzheimer's disease based on the MRI image of brain. However, most of the current technical diagnoses only depend on a single model, such as a convolutional neural network model, and further brain lobe area division and brain lobe area ratio calculation are realized through the single model, so that misjudgment is extremely easy to occur due to the differences of different brains, MRI shooting environments and MRI equipment, and the single model is difficult to adapt to all MRI images, so that misjudgment of Alzheimer's disease based on the MRI images is caused.

Disclosure of Invention

The invention provides a multi-classification method, a multi-classification device and a multi-classification computer-readable storage medium for MRI images of Alzheimer's disease, and mainly aims to solve the problem that the error judgment rate of Alzheimer's disease is high because most methods only depend on a single model to identify brain lobe areas of the MRI images of the brain at present.

In order to achieve the above object, the present invention provides a multi-classification method for MRI images of alzheimer's disease, comprising:

receiving a brain MRI image of a patient, and performing binarization processing on the brain MRI image to obtain a brain binary image, wherein the brain binary image consists of a two-dimensional image matrix with pixel values of 0 or 1;

And sequentially executing segmentation on the two-dimensional image matrix by utilizing a pre-constructed segmentation standard to obtain a plurality of groups of two-dimensional segmentation matrixes with different dimensions, wherein the segmentation standard is as follows: according to the dimension of the two-dimensional blocking matrix, the two-dimensional blocking matrix is sequentially 2 ⁱ *2 ⁱ Performing a segmentation, wherein i = 0,1,..n, n is the dimension of the two-dimensional image matrix;

selecting a blocking matrix with a pixel value of 1 from a plurality of groups of two-dimensional blocking matrices with different dimensions to obtain one or more groups of single-pixel blocking matrices;

determining the blocking positions of each group of single-pixel blocking matrixes in the two-dimensional blocking matrixes and the number of the included pixel values which are 1, and fitting by taking the blocking positions as independent variables and the total number of the pixel values which are 1 as dependent variables to obtain a functional relation;

re-executing binarization processing on the brain MRI image according to the functional relation to obtain an optimized binary image, and executing brain structure splitting on the optimized binary image to obtain a binary temporal lobe map, a binary frontal lobe map, a binary top lobe map and a binary occipital lobe map;

inputting the brain MRI image into a pre-constructed brain structure identification model to obtain an MRI temporal lobe map, an MRI frontal lobe map, an MRI top lobe map and an MRI occipital lobe map, wherein the brain structure automatic identification model consists of a convolution layer, a feature fusion layer, an attention layer and a YOLO model layer;

Calculating the area errors of each brain lobe area between the binary temporal lobe map, the binary frontal lobe map, the binary top lobe map and the binary occipital lobe map and the corresponding MRI temporal lobe map, MRI frontal lobe map, MRI top lobe map and MRI occipital lobe map respectively to obtain temporal lobe area errors, frontal lobe area errors, top lobe area errors and occipital area errors;

if a brain lobe area with the area error larger than a preset area threshold exists, sending the brain MRI image to a doctor to execute the human judgment of the Alzheimer's disease;

if no brain lobe area with the area error larger than the area threshold exists, calculating the area occupation ratio of the MRI temporal lobe image, the MRI frontal lobe image, the MRI top lobe image and the MRI occipital lobe image in the brain MRI image respectively, and determining the disease severity level of the Alzheimer disease of the patient according to the four groups of area occupation ratios.

Optionally, the determining the blocking position of each group of single-pixel blocking matrices in the two-dimensional blocking matrix includes:

constructing a two-dimensional rectangular coordinate system, and projecting the two-dimensional block matrix into the two-dimensional rectangular coordinate system, wherein the lower left corner of the two-dimensional block matrix is the origin of the two-dimensional rectangular coordinate system, and the distance division of the horizontal coordinate and the vertical coordinate of the two-dimensional rectangular coordinate system takes the pixel spacing of the two-dimensional block matrix as the division basis;

And determining the blocking positions of each group of single-pixel blocking matrixes in a two-dimensional rectangular coordinate system in sequence, wherein the blocking positions consist of the central coordinates of the centers of the single-pixel blocking matrixes and the side lengths of the matrixes.

Optionally, the fitting with the block positions as independent variables and the total number of the pixel values of 1 as the dependent variables to obtain a functional relationship includes:

setting all single pixel block matrixesThe corresponding set of matrix center coordinates is C, where c= { (x) _j ，y _j )|j＝0，1，...，m}，(x _j ，y _j ) The center coordinate of the jth matrix center is represented, and m is the total number of matrix centers;

and obtaining the total number of matrix side lengths and pixel values of 1 under each matrix center, and fitting the total number of the matrix side lengths and the pixel values of 1 corresponding to each matrix center by taking the matrix side lengths as independent variables and the total number of the pixel values of 1 as dependent variables respectively.

Optionally, the re-performing binarization processing on the brain MRI image according to the functional relation to obtain an optimized binary image, including:

each center coordinate (x _j ,y _j ) All pixels in the neighborhood are obtained to obtain a neighborhood pixel set, and the following operation is carried out on each neighborhood pixel in the neighborhood pixel set:

obtaining center coordinates (x) _j ,y _j ) Functional relation z of (2) _j =f (l), where z _j The j-th central coordinate is represented by a dependent variable of the total number of pixel points equal to 1, and l represents the side length of the matrix and is an independent variable;

solving a first derivative in the function relation to obtain a first derivative function;

setting the side length of a matrix in the first-order derivative function as 2, and solving to obtain a function value of the first-order derivative function, wherein the function value is called a binarization weight value;

multiplying the binarization weight value with the pixel value of each neighborhood pixel, and judging the size relation between the multiplied pixel value of the neighborhood pixel and 128;

when the pixel value of the multiplied neighborhood pixel is greater than or equal to 128, mapping the pixel value to be equal to 1;

when the pixel value of the multiplied neighborhood pixel is smaller than 128, mapping the pixel value to be equal to 0;

and summarizing each neighborhood pixel which has executed the binarization operation to obtain the optimized binary image.

Optionally, inputting the brain MRI image into a pre-constructed brain structure recognition model to obtain an MRI temporal lobe map, an MRI frontal lobe map, an MRI top lobe map and an MRI occipital lobe map, including:

performing image enhancement on the brain MRI image to obtain a brain enhanced image set

Performing convolution operation on the brain MRI image and the brain enhancement image set respectively by using the convolution layers to obtain a brain convolution image and an enhancement convolution image set respectively, wherein the convolution operation of the convolution layers sequentially uses convolution kernels 2 x 2, 5*5 and 10 x 10, and the step length of the convolution kernels is 1;

Performing activation processing on the brain convolution image and the enhanced convolution image set by utilizing a RELU function to obtain a brain activation image and an enhanced activation image set;

in the feature fusion layer, the brain activated image and the enhanced activated image set are added in sequence according to the pixel correspondence principle to obtain a brain fusion image;

inputting the brain fusion image and the brain convolution image into an attention layer to execute attention calculation, so as to obtain brain MRI characteristics;

and performing brain structure identification on the brain MRI features by utilizing the YOLO model layer to obtain an MRI temporal lobe map, an MRI frontal lobe map, an MRI top lobe map and an MRI occipital lobe map.

Optionally, the image enhancement includes shearing, rotation, scaling, translation, scale, contrast, noise perturbation.

Optionally, the inputting the brain fusion image and the brain convolution image into an attention layer performs attention calculation to obtain brain MRI features, including:

performing convolution operation with the step length of 2 and the convolution kernel of 2 x 2 on the brain fusion image and the brain convolution image to respectively obtain a brain fusion image and a brain convolution image which are subjected to convolution;

and acquiring weight values and offset values of the attention layer for performing attention calculation, and performing the attention calculation on the brain fusion image and the brain convolution image which are subjected to convolution according to the weight values and the offset values to obtain the brain MRI characteristics.

Optionally, the performing attention computation on the convolved brain fusion image and the brain convolution image according to the weight value and the offset value to obtain the brain MRI feature includes:

the brain MRI features were calculated using the following attention calculation formula:

h＝f ₂ (f ₁ (W _t t+W _g g)+b)

wherein h represents the brain MRI feature, f ₁ And f ₂ For two groups of preset activation functions in the attention layer, t is a brain fusion image with complete convolution, g is a brain convolution image with complete convolution, W _t Weight value of t, W _g The weight value of g, and b is the bias value.

In order to solve the above problems, the present invention also provides an MRI image multi-classification apparatus for alzheimer's disease, the apparatus comprising:

the brain binary image comprises a two-dimensional image matrix segmentation module, a two-dimensional image matrix segmentation module and a brain binary image processing module, wherein the two-dimensional image matrix segmentation module is used for receiving a brain MRI image of a patient, performing binarization processing on the brain MRI image to obtain a brain binary image, the brain binary image comprises a two-dimensional image matrix with a pixel value of 0 or 1, and sequentially performing segmentation on the two-dimensional image matrix by utilizing a pre-constructed segmentation standard to obtain a plurality of groups of two-dimensional segmentation matrixes with different dimensions, wherein the segmentation standard is as follows: according to the dimension of the two-dimensional blocking matrix, the two-dimensional blocking matrix is sequentially 2 ⁱ *2 ⁱ Performing a segmentation, wherein i = 0,1,..n, n is the dimension of the two-dimensional image matrix;

the function fitting module is used for selecting a blocking matrix with a pixel value of 1 from a plurality of groups of two-dimensional blocking matrixes with different dimensions to obtain one or a plurality of groups of single-pixel blocking matrixes, determining the blocking position of each group of single-pixel blocking matrixes in the two-dimensional blocking matrix and the number of the included pixel values of 1, fitting the blocking position as an independent variable and the total number of the pixel values of 1 as a dependent variable to obtain a functional relation, re-executing binarization processing on the brain MRI image according to the functional relation to obtain an optimized binary image, and executing brain structure splitting on the optimized binary image to obtain a binary temporal lobe map, a binary frontal lobe map, a binary parietal lobe map and a binary occipital map;

the brain structure identification module is used for inputting the brain MRI image into a pre-constructed brain structure identification model to obtain an MRI temporal lobe map, an MRI frontal lobe map, an MRI top lobe map and an MRI occipital lobe map, wherein the brain structure automatic identification model consists of a convolution layer, a feature fusion layer, an attention layer and a YOLO model layer;

the disease weight grade judging module is used for respectively calculating the area errors of the binary temporal lobe diagram, the binary frontal lobe diagram, the binary top lobe diagram and the binary occipital diagram and the corresponding brain lobe areas among the MRI temporal lobe diagram, the MRI frontal lobe diagram, the MRI top lobe diagram and the MRI occipital diagram to obtain temporal lobe area errors, frontal lobe area errors, top lobe area errors and occipital area errors, if the brain lobe areas with the area errors being larger than a preset area threshold exist, the brain MRI image is sent to a doctor to execute the artificial judgment of the Alzheimer's disease, if the brain lobe areas with the area errors being larger than the area threshold do not exist, the area occupation ratios of the MRI temporal lobe diagram, the MRI frontal lobe diagram, the MRI top lobe diagram and the MRI occipital diagram in the brain MRI images are calculated, and the disease weight grade of the Alzheimer's disease of the patient is determined according to the four groups of area occupation ratios.

In order to solve the above-mentioned problems, the present invention also provides an electronic apparatus including:

at least one processor; the method comprises the steps of,

a memory communicatively coupled to the at least one processor; wherein,

the memory stores instructions executable by the at least one processor to implement the multi-classification method of MRI images of alzheimer's disease described above.

In order to solve the above-mentioned problems, the present invention also provides a computer-readable storage medium having stored therein at least one instruction that is executed by a processor in an electronic device to implement the above-mentioned multi-classification method of MRI images of alzheimer's disease.

Compared with the problems described in the background art, the embodiment of the invention receives the brain MRI image of a patient, carries out binarization processing on the brain MRI image to obtain a brain binary image, the most main function of the brain binary image is to construct a two-dimensional image matrix, the two-dimensional image matrix is to conveniently extract a block matrix with 1 pixel value from the two-dimensional image matrix, wherein the block matrix with 1 pixel value is called a single pixel block matrix, in order to improve the identification accuracy of each brain lobe region of the brain, the embodiment of the invention determines the block position of each group of the single pixel block matrix in the two-dimensional block matrix and the number of the included pixel values as 1, uses the block position as an independent variable, uses the total number of the pixel values as 1 as a dependent variable to obtain a functional relation, and uses the functional relation to carry out binarization processing on the brain MRI image again to obtain an optimized binary image, and carries out brain structure splitting on the optimized binary image to obtain a binary lobe map, binary lobe map and binary lobe map, in order to improve the identification accuracy of each brain lobe region, the embodiment of the invention is not used for simply identifying the depth of each brain lobe region, compared with the MRI image, the MRI image is calculated, the MRI image is prevented from being consumed by the optimal lobe region, and the MRI image is calculated, compared with the MRI image is prevented from being required to be compared with the region of the MRI image, and the region of the MRI image is calculated, and the region is required to be optimized by the region and the MRI image, the area errors of the brain lobe areas among the MRI frontal lobe image, the MRI top lobe image and the MRI occipital lobe image are obtained, namely if the area errors are larger than brain lobe areas with preset area thresholds, the brain structure identification model constructed based on deep learning is not necessarily applicable to the brain MRI image of the time, because the brain lobe areas identified by the optimized binarization and the brain structure identification model are too large, if the brain lobe areas with the area errors larger than the area thresholds are not present, the area duty ratio of the MRI temporal lobe image, the MRI frontal lobe image, the MRI top lobe image and the MRI occipital image in the brain MRI image is calculated, and the sick grade of the Alzheimer disease of a patient is determined according to the four groups of area duty ratios, and the brain She Oushi with the area errors larger than the area thresholds is not found, so that the brain lobe areas identified by the optimized binarization and the brain structure identification model have unified results, and can be used for subsequent Alzheimer disease diagnosis. Therefore, the multi-classification method, the multi-classification device, the multi-classification electronic equipment and the multi-classification computer readable storage medium for the MRI images of the Alzheimer's disease mainly aim to solve the problem that the misjudgment rate of the Alzheimer's disease is high because most methods only depend on a single model to identify brain lobe areas of the MRI images of the brain at present.

Drawings

FIG. 1 is a flow chart of a multi-classification method for MRI images of Alzheimer's disease according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a brain lobe structure identified by conventional binarization according to an embodiment of the present invention;

FIG. 3 is a schematic diagram of brain lobe structure identified by the optimized binarization method according to an embodiment of the present invention;

FIG. 4 is a schematic diagram of the position of each lobe in the brain according to an embodiment of the present invention;

FIG. 5 is a functional block diagram of an MRI image multi-classification device for Alzheimer's disease according to an embodiment of the present invention;

fig. 6 is a schematic structural diagram of an electronic device for implementing the multi-classification method of MRI images of alzheimer's disease according to an embodiment of the present invention.

The achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.

Detailed Description

It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

The embodiment of the application provides an MRI image multi-classification method for Alzheimer's disease. The main execution body of the multi-classification method of the Alzheimer's disease MRI image comprises, but is not limited to, at least one of a server, a terminal and the like which can be configured to execute the method provided by the embodiment of the application. In other words, the multi-classification method of the MRI image of the alzheimer's disease may be performed by software or hardware installed at a terminal device or a server device. The service end includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, and the like.

Example 1:

referring to fig. 1, a flow chart of an MRI image multi-classification method for alzheimer's disease according to an embodiment of the present invention is shown. In this embodiment, the multi-classification method for MRI images of alzheimer's disease includes:

s1, receiving a brain MRI image, and performing binarization processing on the brain MRI image to obtain a brain binary image, wherein the brain binary image consists of a two-dimensional image matrix with pixel values of 0 or 1.

It should be explained that the binarization process is an operation of mapping the original pixel value of the brain MRI image to a pixel value of 0 or 1, and the main purpose is to quickly find the edge of the brain and segment the brain structure. For example, if the brain MRI image is an off-white image, the range of variation of the original pixel value is between [0,255], and thus all pixel values of [0,255] are changed to 0 or 1. The conventional binarization mapping is to take the intermediate value, that is, the pixel value smaller than 128 is changed into 0, and the pixel value larger than or equal to 128 is changed into 1, but the binarization mapping does not consider the image characteristics of the brain MRI image, and is extremely easy to cause the phenomenon of important information loss, and referring to the conventional binarization method shown in fig. 2, it can be seen that although the brain edge and the brain internal structure can be roughly identified, information loss still exists. Therefore, the embodiment of the invention reconstructs a binarization method to improve the recognition effect of the edges of the brain and the split brain structure.

S2, sequentially executing segmentation on the two-dimensional image matrix by utilizing a pre-built segmentation standard to obtain a plurality of groups of two-dimensional segmentation matrices with different dimensions.

It should be explained that the partitioning criteria are: according to the dimension of the two-dimensional blocking matrix, the two-dimensional blocking matrix is sequentially 2 ⁱ *2 ⁱ Segmentation is performed, where i=0, 1.

For example, if the two-dimensional image matrix is 1000×1000, that is, n is 1000, so that the two-dimensional image matrix of 1000×1000 is divided according to 2×2, a total of 500 sets of 2×2 two-dimensional block matrices are obtained, then the two-dimensional image matrix of 1000×1000 is divided according to 4*4, a total of 250 sets of two-dimensional block matrices are obtained, and so on, to obtain a plurality of sets of two-dimensional block matrices with different dimensions.

S3, selecting a blocking matrix with a pixel value of 1 from a plurality of groups of two-dimensional blocking matrixes with different dimensions to obtain one or more groups of single-pixel blocking matrixes.

As can be seen from the above, the two-dimensional image matrix is split into a plurality of sets of two-dimensional block matrices with different dimensions through the splitting operation, and therefore, each two-dimensional block matrix may not include a pixel with a pixel value of 1, and may include a pixel with a pixel value of 1, where more than one pixel with a pixel value of 1 is included. Therefore, the embodiment of the invention screens out all the two-dimensional block matrixes with the pixel value of 1 to obtain one or more groups of single-pixel block matrixes.

S4, determining the blocking positions of each group of single-pixel blocking matrixes in the two-dimensional blocking matrixes and the number of the included pixel values which are 1, and fitting by taking the blocking positions as independent variables and the total number of the pixel values which are 1 as the dependent variables to obtain a functional relation.

It should be explained that each group of single-pixel block matrixes has corresponding block positions in the two-dimensional block matrixes, wherein the block positions are determined by the matrix center and the side length of each group of single-pixel block matrixes. In detail, the determining the blocking position of each group of single-pixel blocking matrix in the two-dimensional blocking matrix includes:

For example, if the two-dimensional image matrix is 1000×1000, the length of the abscissa and the length of the ordinate of the two-dimensional rectangular coordinate system are both 1000, and assuming that there is a single-pixel block matrix of 2×2 in the lower left corner of the two-dimensional image matrix, the side length of the single-pixel block matrix is 2, and the center coordinate of the matrix is (1, 1). It can be seen that there are 800 sets of single-pixel block matrices, where the first set of single-pixel block matrices has a side length of 2, a center coordinate of (1, 1), a total number of pixel values of 1 is 1, the second set of single-pixel block matrices has a side length of 4, a center coordinate of (2, 2), a total number of pixel values of 1 is 3, and so on.

In detail, the fitting with the block positions as independent variables and the total number of pixel values of 1 as the dependent variables to obtain a functional relationship includes:

setting a matrix center coordinate set corresponding to all single pixel block matrixes as C, wherein C= { (x) _j ，y _j )|j＝0，1，...，m}，(x _j ，y _j ) The center coordinate of the jth matrix center is represented, and m is the total number of matrix centers;

and obtaining the total number of matrix side lengths and pixel values of 1 under each matrix center, and respectively fitting to obtain the functional relation of the matrix side lengths and the total number of pixel values of 1 corresponding to each matrix center by taking the matrix side lengths as self-organized quantity and the total number of pixel values of 1 as dependent variables.

For example, if the center coordinates of a part of the single-pixel block matrices overlap, and only 200 sets of center coordinates including (1, 1), (2, 2), (4, 4) and the like are present, the matrix side lengths of all the single-pixel block matrices with (1, 1) as the center coordinates are sequentially obtained, and if 2, 4, 16 and the like are present, and the total number of the corresponding pixel values is 1, 2, 5 and the like, it is seen that a functional relationship between the matrix side lengths and the total number of the pixel values is 1 can be obtained by fitting such a relationship. It should be explained that, software such as MATLAB, python or a program library fitting may be used, and if there are multiple groups of single-pixel block matrices corresponding to the total number of different pixel values of 1 under the condition that the matrix side lengths are the same, the average is removed.

As can be seen from the above description, through a series of fitting, different functional relationships can be established under different central coordinates, where each functional relationship uses a matrix side length as an independent variable, and the total number of pixel values is 1 as a dependent variable, and when the central coordinates are (1, 1), the functional relationship between the corresponding matrix side length and the total number of pixel values is 1 may be a unitary primary relationship, and the functional relationship between the matrix side length corresponding to the central coordinates are (2, 2) and the total number of pixel values is 1 may be an exponential relationship.

S5, re-executing binarization processing on the brain MRI image according to the functional relation to obtain an optimized binary image, and executing brain structure splitting on the optimized binary image to obtain a binary temporal lobe map, a binary frontal lobe map, a binary top lobe map and a binary occipital lobe map.

According to the above S1, the conventional binarization is to change the pixel value smaller than 128 to 0 and the pixel value larger than or equal to 128 to 1, which is liable to cause important information loss because the image characteristics of the brain MRI image are not considered. Therefore, the embodiment of the invention constructs the functional relation between the side length of the matrix and the total number of the pixel values of 1, and executes binarization processing on the matrix based on the functional relation so as to reserve more brain information and complete the segmentation of the brain edge and the brain structure with higher precision. In detail, the re-executing binarization processing on the brain MRI image according to the functional relation to obtain an optimized binary image, including:

Each center coordinate (x _j ，y _j ) All pixels in the neighborhood are obtained to obtain a neighborhood pixel set, and the following operation is carried out on each neighborhood pixel in the neighborhood pixel set:

obtaining center coordinates (x) _j ，y _j ) Functional relation z of (2) _j =f (l), where z _j The j-th central coordinate is represented by a dependent variable of the total number of pixel points equal to 1, and l represents the side length of the matrix and is an independent variable;

For example, if the center coordinate is (1, 1), it corresponds to 4 sets of adjacent pixels, respectively (1, 0), (0, 1), (2, 1), (1, 2), and assuming that the center coordinate is (1, 1) as a function of:

z _j ＝al ² +b

Then the first order derivative of the function relationship is solved to obtain a first order derivative function as follows:

z _j ′＝2al

therefore, setting l=2, the value of the first order derivative function can be calculated, namely, the value is a binarization weight value, and according to the multiplication and judgment operation, the binarization operation is re-executed, so that the optimized binary image can be obtained.

It should be emphasized that, in the process of generating the optimized binary image, the pixel value relationship between the adjacent pixels is considered, so that detail information is not easy to lose, so that the segmentation of the brain edge and the brain structure is more accurate than the traditional binarization, and the brain structure is more accurately segmented and more beneficial to the subsequent judgment of the Alzheimer's disease by comparing the graph of FIG. 2 with the optimized binary image shown in FIG. 3.

It is emphasized that the manifestations of Alzheimer's disease can be classified into three classes of light, moderate and severe. The MRI image of the patient with mild senile dementia can find that the temporal lobe area has mild atrophy, and other brain lobe areas cannot be obvious; for patients with moderate senile dementia, atrophy of the whole brain lobe area can be seen clinically, wherein the atrophy of the temporal lobe area is relatively heavy, and the brain lobe atrophy of patients with severe senile dementia is more obvious by analogy. Therefore, according to the optimized binary image obtained after binarization, the embodiment of the invention performs brain structure splitting on the optimized binary image to obtain a binary temporal lobe map, a binary frontal lobe map, a binary top lobe map and a binary occipital lobe map, which mainly aims at calculating atrophy conditions of each brain lobe area, and referring to fig. 4 about distribution diagram of each brain lobe in brain.

However, it will be appreciated that due to MRI imaging environment, human brain structure, instrument variability, etc., the resolution of the imaged brain MRI image by only one method may not be accurate, and thus the embodiment of the present invention introduces a second method: and (3) a model of deep learning, namely automatically identifying each brain lobe region of the brain MRI image, comparing the temporal lobe, the frontal lobe, the parietal lobe and the occipital lobe which are respectively identified by the two methods, if the difference is smaller than a preset standard, indicating that the brain lobe region is accurately segmented, directly calculating the atrophy proportion, and if the difference is larger than the preset standard, manually performing identification to prevent misdiagnosis.

S6, inputting the brain MRI image into a pre-constructed brain structure recognition model to obtain an MRI temporal lobe map, an MRI frontal lobe map, an MRI top lobe map and an MRI occipital lobe map, wherein the brain structure automatic recognition model consists of a convolution layer, a feature fusion layer, an attention layer and a YOLO model layer.

It should be explained that the brain structure recognition model is constructed based on a deep learning model, and model training needs to be performed on the brain structure recognition model before the brain structure recognition model is used, and the training process is not significantly different from other deep learning models, which is not described herein.

However, it should be emphasized that, in order to improve accuracy of brain structure recognition, the embodiment of the present invention optimizes an automatic brain structure recognition model, which mainly includes a convolution layer, a feature fusion layer, an attention layer, and a YOLO model layer. In detail, the inputting the brain MRI image into a pre-constructed brain structure recognition model to obtain an MRI temporal lobe map, an MRI frontal lobe map, an MRI top lobe map and an MRI occipital lobe map includes:

In embodiments of the present invention, the image enhancement includes, but is not limited to, clipping, rotation/reflection/flip transforms, scaling transforms, translation transforms, scale transforms, contrast transforms, noise perturbations, color transforms, and the like.

Further, the inputting the brain fusion image and the brain convolution image into an attention layer to execute attention calculation, to obtain brain MRI features, includes:

In detail, the performing attention computation on the convolved brain fusion image and the brain convolution image according to the weight value and the offset value to obtain the brain MRI feature comprises the following steps:

h＝f ₂ (f ₁ (W _t t+W _g g)+b)

Furthermore, the YOLO model layer is a brain structure recognition layer mainly constructed through a YOLO model, the YOLO model is a disclosed high-quality target detection model, and when the embodiment of the invention inputs the brain MRI features obtained by convolution and self-care calculation into the YOLO model layer, an MRI temporal lobe map, an MRI frontal lobe map, an MRI top lobe map and an MRI occipital map can be recognized from a brain MRI image through the YOLO model.

And S7, calculating the area errors of each brain lobe area between the binary temporal lobe map, the binary frontal lobe map, the binary top lobe map and the binary occipital map and the corresponding MRI temporal lobe map, MRI frontal lobe map, MRI top lobe map and MRI occipital map respectively to obtain temporal lobe area errors, frontal lobe area errors, top lobe area errors and occipital area errors.

It can be understood that, by the improved binarization method according to the embodiment of the present invention, a binary temporal lobe map, a binary frontal lobe map, a binary top lobe map, and a binary occipital map can be identified, and in addition, an MRI temporal lobe map, an MRI frontal lobe map, an MRI top lobe map, and an MRI occipital map are also identified by the brain structure identification model, so as to prevent a large error between the two methods for automatically identifying brain lobe areas, and therefore, a temporal lobe area error, a frontal lobe area error, a top lobe area error, and an occipital area error are calculated respectively according to S7.

And S8, if a brain lobe region with the area error being larger than a preset area threshold exists, sending the brain MRI image to a doctor for executing the human judgment of the Alzheimer' S disease.

It can be understood that if there is a brain lobe region with an area error greater than the preset area threshold, it means that there is a difference in technical recognition between the two methods for region division of the brain MRI image, so if the patient's severity level of the alzheimer's disease is still automatically determined, a problem of diagnosis error is very easily caused, so that the brain MRI image is sent to a doctor to perform human determination of the alzheimer's disease.

And S9, if no brain lobe region with the area error larger than the area threshold exists, calculating the area occupation ratio of the MRI temporal lobe image, the MRI frontal lobe image, the MRI top lobe image and the MRI occipital lobe image in the brain MRI image respectively, and determining the disease severity level of the Alzheimer disease of the patient according to the four groups of area occupation ratios.

When the optimized binarization and brain structure recognition model is within a controllable range for brain lobe region recognition errors, the technical recognition difference of the two methods is smaller, and in order to improve diagnosis efficiency of the disease severity level of the Alzheimer's disease, the area occupation ratio of an MRI temporal lobe image, an MRI frontal lobe image, an MRI top lobe image and an MRI occipital lobe image in a brain MRI image is calculated in sequence, whether atrophy conditions exist in temporal lobes, frontal lobes, top lobes and occipital lobes can be judged according to the condition of the area occupation ratio, and whether the Alzheimer's disease of a patient is primary, intermediate or heavy is determined according to the atrophy proportion.

Example 2:

fig. 5 is a functional block diagram of an MRI image multi-classification device for alzheimer's disease according to an embodiment of the present invention.

The multi-classification device 100 for MRI images of alzheimer's disease according to the present invention may be mounted in an electronic device. Depending on the functions implemented, the multi-classification device 100 for MRI images of alzheimer's disease may include a two-dimensional image matrix segmentation module 101, a function fitting module 102, a brain structure identification module 103, and a disease severity level determination module 104. The module of the invention, which may also be referred to as a unit, refers to a series of computer program segments, which are stored in the memory of the electronic device, capable of being executed by the processor of the electronic device and of performing a fixed function.

The two-dimensional image matrix segmentation module 101 is configured to receive a brain MRI image of a patient, perform binarization processing on the brain MRI image to obtain a brain binary image, where the brain binary image is composed of a two-dimensional image matrix including pixel values of 0 or 1, and sequentially segment the two-dimensional image matrix by using a pre-constructed segmentation standard to obtain a plurality of groups of two-dimensional segmentation matrices with different dimensions, where the segmentation standard is: according to the dimension of the two-dimensional blocking matrix, the two-dimensional blocking matrix is sequentially 2 ⁱ *2 ⁱ Performing a segmentation, wherein i = 0,1,..n, n is the dimension of the two-dimensional image matrix;

the function fitting module 102 is configured to select a block matrix containing pixel values of 1 from a plurality of groups of two-dimensional block matrices with different dimensions to obtain one or more groups of single-pixel block matrices, determine a block position of each group of single-pixel block matrices in the two-dimensional block matrix and the number of the included pixel values of 1, fit a total number of the pixel values of 1 as dependent variables with the block position as an independent variable to obtain a functional relationship, re-perform binarization processing on the brain MRI image according to the functional relationship to obtain an optimized binary image, and perform brain structure splitting on the optimized binary image to obtain a binary temporal lobe map, a binary frontal lobe map, a binary parietal lobe map and a binary occipital map;

the brain structure recognition module 103 is configured to input the brain MRI image into a pre-constructed brain structure recognition model to obtain an MRI temporal lobe map, an MRI frontal lobe map, an MRI top lobe map, and an MRI occipital lobe map, where the brain structure automatic recognition model is composed of a convolution layer, a feature fusion layer, an attention layer, and a YOLO model layer;

the disease severity judging module 104 is configured to calculate the area errors of each lobe area between the binary temporal lobe map, the binary frontal lobe map, the binary top lobe map, and the binary occipital map and the corresponding MRI temporal lobe map, MRI frontal lobe map, MRI top lobe map, and MRI occipital map, obtain temporal lobe area errors, frontal lobe area errors, top lobe area errors, and occipital area errors, if there is a lobe area with an area error greater than a preset area threshold, send the brain MRI image to a doctor to perform human judgment of alzheimer's disease, if there is no lobe area with an area error greater than the area threshold, calculate the area occupation ratio of the MRI temporal lobe map, MRI frontal lobe map, MRI top lobe map, and MRI occipital map in the brain MRI image, and determine the disease severity of the patient according to the four groups of area occupation ratios.

In detail, the modules in the MRI image multi-classification device 100 for alzheimer's disease in the embodiment of the present invention use the same technical means as the MRI image multi-classification method for alzheimer's disease described in fig. 1, and can produce the same technical effects, which are not described herein.

Example 3:

The electronic device 1 may comprise a processor 10, a memory 11, a bus 12 and a communication interface 13, and may further comprise a computer program stored in the memory 11 and executable on the processor 10, such as an Alzheimer's disease MRI image multi-classification program.

The memory 11 includes at least one type of readable storage medium, including flash memory, a mobile hard disk, a multimedia card, a card memory (e.g., SD or DX memory, etc.), a magnetic memory, a magnetic disk, an optical disk, etc. The memory 11 may in some embodiments be an internal storage unit of the electronic device 1, such as a removable hard disk of the electronic device 1. The memory 11 may in other embodiments also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card) or the like, which are provided on the electronic device 1. Further, the memory 11 may also include both an internal storage unit and an external storage device of the electronic device 1. The memory 11 may be used not only for storing application software installed in the electronic device 1 and various types of data, such as codes of multi-classification programs of MRI images of alzheimer's disease, but also for temporarily storing data that has been output or is to be output.

The processor 10 may be comprised of integrated circuits in some embodiments, for example, a single packaged integrated circuit, or may be comprised of multiple integrated circuits packaged with the same or different functions, including one or more central processing units (Central Processing unit, CPU), microprocessors, digital processing chips, graphics processors, combinations of various control chips, and the like. The processor 10 is a Control Unit (Control Unit) of the electronic device, connects respective components of the entire electronic device using various interfaces and lines, executes programs or modules (for example, an alzheimer's disease MRI image multi-classification program or the like) stored in the memory 11 by running or executing the programs or modules, and invokes data stored in the memory 11 to perform various functions of the electronic device 1 and process the data.

The bus may be a peripheral component interconnect standard (peripheral component interconnect, PCI) bus or an extended industry standard architecture (extended industry standard architecture, EISA) bus, among others. The bus may be classified as an address bus, a data bus, a control bus, etc. The bus is arranged to enable a connection communication between the memory 11 and at least one processor 10 etc.

Fig. 6 shows only an electronic device with components, it being understood by a person skilled in the art that the structure shown in fig. 6 does not constitute a limitation of the electronic device 1, and may comprise fewer or more components than shown, or may combine certain components, or may be arranged in different components.

For example, although not shown, the electronic device 1 may further include a power source (such as a battery) for supplying power to each component, and preferably, the power source may be logically connected to the at least one processor 10 through a power management device, so that functions of charge management, discharge management, power consumption management, and the like are implemented through the power management device. The power supply may also include one or more of any of a direct current or alternating current power supply, recharging device, power failure detection circuit, power converter or inverter, power status indicator, etc. The electronic device 1 may further include various sensors, bluetooth modules, wi-Fi modules, etc., which will not be described herein.

Further, the electronic device 1 may also comprise a network interface, optionally the network interface may comprise a wired interface and/or a wireless interface (e.g. WI-FI interface, bluetooth interface, etc.), typically used for establishing a communication connection between the electronic device 1 and other electronic devices.

The electronic device 1 may optionally further comprise a user interface, which may be a Display, an input unit, such as a Keyboard (Keyboard), or a standard wired interface, a wireless interface. Alternatively, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch, or the like. The display may also be referred to as a display screen or display unit, as appropriate, for displaying information processed in the electronic device 1 and for displaying a visual user interface.

It should be understood that the embodiments described are for illustrative purposes only and are not limited to this configuration in the scope of the patent application.

The multi-classification program of the MRI image of alzheimer's disease stored in the memory 11 in the electronic device 1 is a combination of instructions which, when run in the processor 10, can realize:

Specifically, the specific implementation method of the above instructions by the processor 10 may refer to descriptions of related steps in the corresponding embodiments of fig. 1 to 3, which are not repeated herein.

Further, the modules/units integrated in the electronic device 1 may be stored in a computer readable storage medium if implemented in the form of software functional units and sold or used as separate products. The computer readable storage medium may be volatile or nonvolatile. For example, the computer readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a U disk, a removable hard disk, a magnetic disk, an optical disk, a computer Memory, a Read-Only Memory (ROM).

The present invention also provides a computer readable storage medium storing a computer program which, when executed by a processor of an electronic device, can implement:

The modules described as separate components may or may not be physically separate, and components shown as modules may or may not be physical units, may be located in one place, or may be distributed over multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional module in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units can be realized in a form of hardware or a form of hardware and a form of software functional modules.

It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof.

Finally, it should be noted that the above-mentioned embodiments are merely for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made to the technical solution of the present invention without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A multi-classification method for MRI images of alzheimer's disease, the method comprising:

determining the blocking positions of each group of single-pixel blocking matrixes in the two-dimensional blocking matrixes and the number of the included pixel values which are 1, and fitting by taking the blocking positions as independent variables and the total number of the pixel values which are 1 as dependent variables to obtain a functional relation; the blocking position consists of a central coordinate of a matrix center of the single-pixel blocking matrix and a matrix side length;

and re-executing binarization processing on the brain MRI image according to the functional relation to obtain an optimized binary image, wherein the method comprises the following steps of:

summarizing each neighborhood pixel which has executed the binarization operation to obtain the optimized binary image;

2. The multi-classification method of alzheimer's disease MRI images of claim 1, wherein said determining the tile positions of each set of single pixel tile matrices in a two-dimensional tile matrix comprises:

3. The multi-classification method of MRI images for alzheimer's disease according to claim 2, wherein said fitting a function relationship with the block locations as independent variables and the total number of pixel values of 1 as dependent variables comprises:

4. The multi-classification method of MRI images of alzheimer's disease according to claim 1, wherein said inputting said MRI images of brain into a pre-constructed brain structure recognition model to obtain MRI temporal lobe map, MRI frontal lobe map, MRI top lobe map and MRI occipital lobe map comprises:

5. The multi-classification method of alzheimer's disease MRI images of claim 4, wherein said image enhancement includes shearing, rotation, scaling, translation, scaling, contrast, noise perturbation.

6. The multi-classification method of alzheimer's disease MRI images of claim 4, wherein said inputting said brain fusion image and brain convolution image to an attention layer performs an attention calculation to obtain brain MRI features, comprising:

7. The multi-classification method of alzheimer's disease MRI images of claim 6, wherein said performing attention calculations on the convolved brain fusion image and brain convolution image based on said weight and bias values, resulting in said brain MRI features, comprises:

h＝f ₂ (f ₁ (W _t t+W _g g)+b)

8. An apparatus for multi-classification of MRI images of alzheimer's disease, said apparatus comprising:

the function fitting module is used for selecting a blocking matrix with a pixel value of 1 from a plurality of groups of two-dimensional blocking matrixes with different dimensions to obtain one group or a plurality of groups of single-pixel blocking matrixes, determining the blocking position of each group of single-pixel blocking matrixes in the two-dimensional blocking matrix and the number of the included pixel values of 1, and fitting the total number of the pixel values of 1 as independent variables by taking the blocking position as the independent variable to obtain a function relationship, wherein the blocking position consists of the central coordinate of the matrix center of the single-pixel blocking matrix and the matrix side length; re-executing binarization processing on the brain MRI image according to the functional relation to obtain an optimized binary image, and executing brain structure splitting on the optimized binary image to obtain a binary temporal lobe map, a binary frontal lobe map, a binary top lobe map and a binary occipital lobe map; and re-executing binarization processing on the brain MRI image according to the functional relation to obtain an optimized binary image, wherein the method comprises the following steps of:

9. An electronic device, the electronic device comprising:

at least one processor; the method comprises the steps of,

a memory communicatively coupled to the at least one processor; wherein,

the memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the multi-classification method of an MRI image of alzheimer's disease according to any of the claims 1-7.